Fig. 5: Area under the receiver operating curve variable importance for baseline antibody titers relative to covariates.

Among pediatric participants, in whom both antibodies were measured (n = 1355), estimated gain in achievable AUC is shown on the y-axis for time to RSV illness by 180 days from enrollment based on adding each RSV antibody to a prediction model containing only baseline covariates (age; school/daycare attendance; immunocompromised status). The AUC is measured on a scale of 0-1 with the y-axis starting at 0.50 to represent the ability to predict RSV illness beyond random chance (AUC = 0.50). The measure of center of the error bars represents the estimated difference in AUC comparing the model with the baseline covariates only to the model with the baseline covariates and the RSV antibody. The error bars represent the 2.5 and 97.5 percentile of the bootstrap replicate estimates for this AUC difference. The AUC estimates are the average of all the bootstrap replicate estimates, which are themselves averages within bootstrap datasets of sample splitting over 10 different seeds. Confidence intervals for the variable importance estimates were constructed using 500 bootstrap replicates with sampling accounting for within-household correlation. Abbreviations: AUC area under the receiver operating curve, RSV respiratory syncytial virus.